Beruflich Dokumente
Kultur Dokumente
Jere R. Behrman
University of Pennsylvania
Lori G. Kletzer
University of California, Santa Cruz
Michael S. McPherson
Macalester College
December 1997
The research reported here was supported by grants from the Andrew W. Mellon Foundation and
the National Science Foundation (SED-9115440).
Published in The Annals of the American Academy of Political and Social Science: The Changing
Educational Quality of the Workforce, Robert Zemsky and Peter Cappelli, eds., September 1998,
pp: 12-23.
Abstract
Understanding the economic payoff to human capital investments is very important from
the standpoint both of individuals and of society. Correctly estimating these impacts necessitates
having a well-developed idea of the microeconomic determinants of human behavior. Without
this, empirical analyses of such topics as career choice, college choice, or wage determination will
be flawed. We begin with a discussion of why these choice models are important - using
examples of similar attempts that do not capture sufficient information - and illustrate their
usefulness in a variety of contexts. We also describe the results of our attempts to examine
college choice using microeconomic models.
Microeconomic analyses of educational investment and career choice examine how the
influence of personal and family background characteristics, actual and expected market
conditions, and economic incentives interact to shape individuals' choices about the sequence of
college and career choice at the microeconomic level emphasize the heterogeneity in various
respects (academic aptitude, family resources, and so on) of the populations studied and the
interactions among these varied characteristics and various opportunities and incentives in
mind is longitudinal and sequential. That is, college and career destinations are seen as the result
of an individual moving along a path over time. Choices made along the way are influenced by the
person's characteristics, which themselves are partly a function of past choices and also partly a
function of expectations about future opportunities and alternatives. The facts that these choices
are extended over time and that choices at a given time are responsive to the results of past
choices and to expectations about the future have crucial implications for empirical modeling
First, empirical estimation of such models virtually requires the availability of longitudinal
data on individuals – that is, observations of the choices made by individuals at various stages in
the evolution of their educational and career paths. Second, the complex temporal
empirical analysis of such data. Two particular implications should be underscored. First, it is
clear in sequential modeling that the alternatives available at time t1 are dependent on
1
choices made at an earlier time t0 -- for example, the decision to go to Julliard for college
certainly conditions one's later choice of major. But with this kind of interdependence it
becomes very easy to believe that attendance at Julliard has a causal impact on one's likelihood of
majoring in piano when in fact it may be that a disposition to major in piano produced both
the decision to go to Julliard and the decision to major in piano. In principle, if one could
accurately measure the disposition to major in music at, say, age sixteen, one could
control for this factor in estimating the impact of Julliard attendance on choice of major.
(Equivalently, one could achieve such control by having good measures of all the factors that
determine the disposition at age sixteen.) Unfortunately, one is likely to lack complete measures.
In that case, the omitted variables that influence both the tendency to go to Julliard and
the tendency to major in piano are likely to lead to overestimating the "Julliard" effect on choice
of major.
The second, and related, problem is the influence of expected future opportunities on
choices at a particular time. Here the essential difficulty is posed by the fact that
student, for example, will expect greater returns from pursuit of a mathematics career than would
someone with limited mathematical aptitude. If the researcher possessed a good independent
this dependency could be explicitly modeled and would not be problematic. In the absence of
such measures, there are great difficulties in sorting out the effects on career choice that
operate through influencing earnings expectations from effects that operate in other ways.
The proper analysis of sequential choice in longitudinal data sets remains a difficult
2
analytical and empirical issue. Fortunately, however, the last twenty years have brought real
improvements in our capacity to perform such analyses. Several large longitudinal data sets on
histories and employment histories have been developed with federal support. Over the same
period, improved statistical techniques to cope with the temporal interdependence of choices
Below, we examine as a case study a career with special significance for a nation's
economic future, science and engineering, before turning to a broader discussion of career
choice. We then report more briefly on some original empirical results regarding college choice
and future wages that result from the type of estimation methods suggested here.
Only the foolhardy defend economic models on grounds oftheir realism. Yet the
simplifications inherent in many models of the labor market for scientists and engineers seem
particularly unsettling. Many of these models, including some rather influential ones, are almost
purely mechanical: with fixed proportions of students grinding through high school, some
other fixed proportion grinding through college, some percentage of them going into scientific or
engineering careers, some fraction of others going to graduate school and so on. The
proven ability of these models to be quite spectacularly and consistently wrong has not been
More sophisticated models recognize the interaction of supply and demand in determining
labor market outcomes. Demand increases raise wages and, with a lag, yield an increase in
quantity supplied. While it has proved difficult to come up with stable and believable parameters
3
for such models, they highlight an important adjustment mechanism that is bound to play a role
Yet even these models overlook or oversimplify aspects of the operation of markets for
educated labor that intuitively seem quite important. While the presence of lags in the adjustment
process in aggregate supply-demand models recognizes that some rather complicated behavior is
taking place, which involves learning and the adjustment of plans, the details of that process
choosing, learning and adjusting at center stage. It is illuminating, we think, to view the process
contrast to aggregate supply- demand models, which tend to suppress recognition of differences
in the characteristics of labor market participants, and thus of differential response to changes in
background, aptitude and circumstances and permit the testing of hypotheses about whether these
differences matter.
To motivate the following discussion, consider two stories about how the market for some
group of scientists -- say microbiologists -- responds to variation in the wages and opportunities
in that field. Suppose wages and job opportunities for microbiologists experience a rather abrupt
drop, but one which is widely believed to be likely to persist for some years. Imagine that we can
ask is: how will the relative supply of students to the field of microbiology vary according to the
4
Scenario I is that the most promising students in microbiology have a real passion for the
subject and feel something of a calling to it. In more prosaic terms, we might say that their
relative advantages as microbiologists, compared to their prospects in other fields, are quite high.
In this scenario, most of the top quality students stick with microbiology while the poorer wages
and job prospects scare the less promising students away. Most of the decline in entry of
microbiologists to the profession occurs at the low end in terms of quality, with the result that the
Scenario II reverses this. Top microbiology students have excellent alternatives and can
with relative ease and little sacrifice migrate from microbiology to other fields of study,
such as law or banking. Less capable students, although subject to the same pressures, face less
good alternatives and are, on the margin, less likely to leave the field. As a result, the average
quality of entering microbiologists falls with declining demand. The economic and social
The issue raised by these two stories illustrates one class of questions that is naturally
studied under the rubric of the microeconomics of career choice. In general, models of science
and engineering labor markets, particularly those that aim at forecasting or policy analysis,
completely neglect the kinds of considerations suggested by these stories. Here we raise the
following question: can insights and evidence from the microeconomic analysis of career choice
be used effectively to improve the usefulness of forecasts and policy analyses of markets for
5
models of science and engineering labor markets. In the late 1980s, the National Science
Foundation (NSF) produced forecasts of substantial "shortfalls" of scientists and engineers at both
bachelors and advanced levels through the 1990s and beyond. Although the producers of these
analyses introduced various qualifications into their analyses, policy advocates both inside and
outside NSF relied on these forecasts in arguing for programs that would increase the number
In the last several years, a variety of circumstances have resulted in substantial declines in
the demand for scientists and engineers, leading many observers to conclude that the nation
faces a surplus instead of a shortfall of qualified science and engineering personnel. Congressional
hearings have focused on the inadequacies of the NSF modeling efforts, and substantial efforts
Some of the limitations of existing models lie on the demand side. The controversial NSF
models, for example, measure demand for new scientists and engineers simply as the number of
newly produced scientists and engineers in a benchmark period which is judged to be one of
rough supply-demand balance. Whether or not the number of science and engineering personnel
produced in that year could be taken as a good measure of demand in that year, it is certainly
quite arbitrary to suppose that the same level of production would be suitable in other years. An
adequate model of demand would need to take into account the influence of underlying economic
factors, such as changes in the industrial mix and changes in technology, on demand for scientists
and engineers.
More directly pertinent to issues of career choice are limitations on the treatment of supply
in existing models. The NSF model derives a supply of new college graduates by applying a
6
coefficient to the size of the college-age population. This coefficient is determined judgmentally
on the basis of recent experience; the model does not make the coefficient sensitive to
variation in the supply-demand balance. The yield of scientists and engineers from the college
graduate pool is similarly determined by a judgmental coefficient, which again is not sensitive to
Many other models of science and engineering labor markets share major characteristics
with the NSF model, although some are more elaborate in various ways. Thus, some models
disaggregate fields within science and engineering more finely; some recognize more stages in the
educational "pipeline" – introducing a separate step from college entry to college completion, for
example.
Perhaps the most significant departure from the NSF type of model is found in those
models that explicitly incorporate feedback between supply and demand. Richard Freeman has
contributed a number of studies for various markets for educated labor which explicitly model
lagged wage adjustments and lagged responses of supply and demand to changes in wages as part
of the adjustment process. Other analysts have produced models in which an adjustment of
supply to demand is assumed to occur, but without explicitly incorporating wages or other
It is a controversial question whether these simple aggregate models, whether or not they
incorporate lagged adjustment processes, can capture enough of reality to shed light on likely
future developments in science and engineering labor markets. There cannot, however, be much
doubt that these analyses fall woefully short of allowing analysts to assess the effects of various
policy issues that are brought forward in response to diagnoses of shortages or surpluses in these
7
labor markets.
Consider, for example, the influential analyses of the market for scientists and engineers
published by Richard Atkinson (1990, 1988) almost a decade ago. Atkinson cited the NSF
forecasts of shortfalls in scientists and engineers in making a case for increased support for the
education of people in those fields. Yet both the nature of the problem as he diagnosed it
and the policies he advocated in response to the shortfall assumed knowledge of behavioral
relationships that are not included in typical aggregate models of these labor markets. He
argued, for example, on the basis of evidence which is not integrated into these aggregate models,
that the majority of the interested and qualified students who are "lost" to science and engineering
between their freshman and senior years of college switch to other areas because they find other
fields more promising and believe they would have better job prospects elsewhere. He
emphasized the importance for policy purposes of interesting women and minorities in scientific
and engineering careers, while the aggregate models shed little light on how determinants of the
decision to enter science or engineering differ among rage and gender groups.
Atkinson's policy proposals also presupposed facts about behavioral responses that are not
explored in typical supply- demand models. He proposed in particular that increases in fellowship
opportunities, especially when targeted at women and minorities, would elicit a strong supply
response.
A second set of issues which figures prominently in policy discussions but is largely
absent from modeling efforts is the matter of quality of scientists and engineers. It is commonly
alleged, with at least some empirical support, that the quality of students in undergraduate
engineering programs fluctuates sharply with fluctuations in the state of the engineering labor
8
market. When the market is strong, both the numbers and quality of engineering students rise,
with the rise in quality being due partly to the interest of more able students in engineering and
many observers that fluctuations in the academic quality of persons entering the science and
science, for example, Levin and Stephan (1991) have argued that absence of growth in the
publication productivity of recent cohorts of scientists, despite the fact that they have a larger
knowledge base to work with, may reveal a decline in the quality of the research personnel. They
speculate that such a decline might be related to the relatively weak labor market for academic
The quality issue assumes added importance in policy discussions for two reasons. First, it
seems quite plausible that at the margin adjustments in the employment of scientists and engineers
occur through quality. It is less likely that jobs will go unfilled than that they will be filled by less
qualified personnel -- a fact that has been shown strikingly in the case of high school teachers of
science and mathematics. Second, programs to recruit students to science and engineering are
often very quality oriented, as is the case with fellowship programs of the kind Atkinson
advocates. The issue is generally not only one of attracting enough students or workers, but of
attracting the right students and workers -- and this is an issue which aggregate models of supply
9
We suggest four categories of questions where aggregate models of supply and demand
(1) How large is the response of individuals to changes in the incentives for pursuing
particular career paths? Incentives may include scholarships and fellowships, expected earnings,
expected risks of unemployment and the like. The impact of incentives may vary at different
points in the career path. Once a student has chosen to attend a technical institute like Cal Tech
for college, for example, he or she may be unlikely to become discouraged about pursuing a
science/engineering career. Yet the decision for a student with particular characteristics to attend
such an institution in the first place may well be more strongly influenced by incentives.
(2) How do responses to incentives vary among students with differing characteristics?
Are students with strong mathematical aptitudes more or less sensitive to fluctuation in
(3) How do various social and economic background characteristics influence persons'
(4) What is the interaction between these background characteristics and incentives to
pursue careers in science and engineering? For example, are students from low-income families
more likely to be influenced by fellowship opportunities in their choice of majors and careers? An
affirmative action would provide a reason for aiming to target fellowship money on such
students.
It should be clear enough that aggregate supply and demand models are very limited in
their ability to respond to these questions. Some worthwhile evidence about responses to wages
and labor market conditions can be obtained from time-series models that incorporate response
10
lags. Such studies are bound to be handicapped, however, by the limited number of observations
available in the time-series, and the confounding effects of other economic and social changes that
To push further with aggregate time-series models, by trying to detect effects of changes
changes among different population groups, is almost certainly too ambitious. Convincing
Do we have reason to believe that micro data sets and the kind of longitudinal analysis
described earlier can do better? Plainly the best way to answer this question affirmatively is to
do the analyses and demonstrate their effectiveness. Still, there may be some purpose to trying to
describe the reasons why conceptually this approach seems to hold promise.
The first point to emphasize is that it is very plausible to believe that behavioral responses
vary importantly both at different stages of the life cycle and across groups with different
characteristics. This variation is important in its own right, we have suggested, for policy
purposes, but it also suggests advantages to the use of micro data sets. Time-series
analyses will inevitably require aggregating over people with quite different characteristics and
collapsing different segments of the sequential career path. Such aggregation is certain to
add to the imprecision of estimated behavioral effects and can easily lead to biases.
the extent of behavioral response to, say, expected wages, to vary along the career path
-- affecting the decision to go to college, the choice of major, the decision to complete, and so on
11
in potentially different ways.
Similarly an analysis based on micro data can make explicit the possibility that men and
women or people of different ethnic, social or economic backgrounds may respond differently to
differences in the incentives they face. The opportunity to study interactions between people's
characteristics and their response to variations in expected wages adds an important dimension to
In the following two sections, we report briefly on studies we have undertaken that apply
the kind of sequential microeconomic analysis here to questions about college choice and
There is a large literature concerning college attendance decisions. These decisions are
naturally thought to be influenced strongly by family background. But previous studies of the
impact of family background on college attendance generally suffer from the selectivity problem
we have described previously: if they include high school achievement as a determinant of college
choice, they treat it as predetermined in a statistical sense rather than determined in part by
observed and unobserved components of family background as part of a longer- run educational
strategy. But family background may affect college decisions indirectly through effects on high
school achievement as well as having direct effects. Previous estimates, by treating high school
achievement as predetermined, may be biased and are not informative about the importance of
In Behrman, Kletzer, McPherson and Schapiro (1995), we show statistically that high
school achievement is determined jointly with college investment decisions: that is, student's
12
decisions about working to achieve in high school are influenced by their college ambitions. We
find evidence that high school achievement depends on many of the same background variables
that influence college educational investments and that failure to account for this interdependence
leads one mistakenly to attribute part of the effect of high school achievement to family
background variables that influence achievement levels. Thus, treatment of high school
predetermined has substantial effects on the estimated magnitudes of some responses of interest
and permits the identification of direct versus indirect effects of family background.
among colleges rather than merely on the quantitative nature of college choice (enrollment
in two-year versus four-year schools, for example). School quality variations are large among
four-year colleges and universities and appear to affect post-schooling outcomes. Studies that
focus exclusively on the quantitative aspect of college choices miss the important qualitative
dimension of those choices on which family background may also have direct and indirect
influences.
We estimate the effects of family background on both the qualitative choice whether to
enroll in two-year versus four-year school and, for those who enroll in four-year colleges and
universities, on the qualitative choice among those schools. The inclusion of this qualitative
dimension of choice affects our understanding of the nature and magnitude of family background
effects on college in some important respects. For example, family income has but limited direct
impact on whether students enroll in post-secondary colleges and universities, but has strong
direct effects on the quality of educational institution selected for those who attend private
13
four-year colleges.
This discussion is far from being the final word on the determination of college choice.
Instead, it is intended to indicate that the type of methodology described above can be
usefully applied in analyzing how students select among colleges and universities.
The greatest effort economists have made in studying the impact of college attendance on
the economy has undoubtedly been in terms of estimating the returns to human capital. Again, the
There has been a rising interest in understanding better the impact of college choices on
wages, an interest motivated by concerns about increasing wage inequalities, about increasing
costs of elite private colleges, and about the perceived increasing roles of highly educated
previous studies have been subject to several limitations. For one, they have not considered
possible differential impacts for different demographic groups identified by race and gender. And,
once again, with few exceptions, they have not considered the choice dimensions with regard to
Behrman, Constantine, Kletzer, McPherson, and Schapiro (1996) addresses both of these
limitations. Estimates are obtained of the impact of time in college and of college quality choices
on wages for four demographic groups based on an explicit dynamic framework. These estimates
suggest that treating time in college and college quality as choices within a dynamic framework
affects fairly substantially the estimated effects. Based on these results, most previous studies
may have overestimated the impact of both college quality and time in college because of the
14
failure to deal with such choices and, in a lesser but still substantial number of cases, because of
The estimates also suggest that there are important demographic differences in the wage
benefits of college. The estimated wage benefits from higher college quality and more time
in college tend to be highest for nonwhite males and next for nonwhite females, then white
females and least for white males. For some members of all groups there appear to be some
incentives for increasing college quality (by paying somewhat higher tuition or whatever), more so
for those who are longer in college because of interactions between college quality and time in
college. But these incentives differ strongly across groups, being much higher for nonwhite males
and much lower for white males, with the females in between. For nonwhites there also appear to
be incentives (again stronger for males than females) to increase time in college. For whites, in
contrast, the estimated net gains do not appear to create such incentives (at least at a reasonable
real discount rate of 4%) - a result that contrasts with common interpretations of estimated
earnings equations.
The results for nonwhites raise important questions regarding why there is unrealized
potential for reaping net benefits from more time in college or attending higher quality colleges.
A number of possible answers to this question - poor information about the impact of college,
imperfect capital markets for financing college, poor information about college quality
differentials, discriminatory admissions - may have important policy implications on both equity
Conclusion
Understanding the economic payoff to human capital investments is very important both
15
from the standpoint of individuals and of society. But it is our view that to estimate these impacts
behavior. Without this, empirical analyses of such topics as career choice, college choice, or
wage determination will be flawed. Our intention here has been to lay out our reasoning as to
why these choice models are important and to illustrate their usefulness in a variety of
contexts.
16
Bibliography
Atkinson, Richard C. 1988. "Bold Steps Are Needed to Educate The Next Generation of
Scientists," The Chronicle of Higher Education, March 2, 34(25), p. B1.
Atkinson, Richard C. 1990. "Supply and Demand for Scientists and Engineers: A National Crisis
in the Making," Science, April 27, 248(4954), pp. 425-432.
Behrman, Jere R., Lori G. Kletzer, Michael S. McPherson, Morton Owen Schapiro, 1995. "How
Family Background Sequentially Affects College Choices: High School Achievement,
College Enrollment and College Quality." February.
Behrman, Jere R., Jill Constantine, Lori G. Kletzer, Michael S. McPherson, Morton Owen
Schapiro, 1996. "Impact of College Quality Choices on Wages: Are There Differences
Among Demographic Groups?" July.
Ehrenberg, Ronald G. 1992. "The Flow of New Doctorates," Journal of Economic Literature,
June, 30(2), pp. 830-875.
Freeman, Richard. 1975. "Supply and Salary Adjustments to the Changing Science Manpower
Market: Physics, 1948-1973." American Economic Review, 65(1), pp. 27-39.
Leslie, Larry R., and Ronald L. Oaxaca. 1993. "Scientist and Engineer Supply and Demand," in
Higher Education: Handbook of Theory and Research. Vol IX. Ed.: John C. Smart. New
York: Agathon Press, pp. 154-211.
Levin, Sharon G., and Paula E. Stephan. 1991. "Research Productivity Over the Life Cycle:
Evidence for Academic Scientists," American Economic Review, 81(1), pp. 114-132.
Levy, David M. 1988. "The Market for Fame and Fortune," Historical Political Economy,
Winter, 20(4), pp. 615-625.
National Academy of Sciences. 1995. Reshaping the Graduate Education of Scientists and
Engineers. Washington, DC: National Academy Press.
National Research Council. 1979. Research Excellence Through the Year 2000: The Importance
of Maintaining a Flow of New Faculty into Academic Research, Washington, DC:
National Academy of Sciences.
17
National Science Board. 1993. Science and Engineering Indicators-1993. Washington, DC: U.S.
Government Printing Office.
National Science Foundation. 1994. Characteristics of Doctoral Scientists and Engineers in the
United States: 1991. NSF 94-307. Arlington, VA.
National Science Foundation. 1988. Doctoral Scientists and Engineers: A Decade of Change,
Washington, DC: National Science Foundation, 88-302.
Stapleton, David C. 1989. "Cohort Size and the Academic Labor Market," Journal of Human
Resources, Spring, 24(2), pp. 221-252.
18